REPO PACKAGE
lllllllama rigorpilot-skills
COMMUNITY11skills
11ready to use
LABSCO SUMMARY
The reproduction lane is README-first and trusted by default: repo-intake-and-plan reads the repository and proposes the smallest documented command to run, env-and-assets-bootstrap prepares a conservative environment, minimal-run-and-audit and run-train execute and record evidence, paper-context-resolver fills a narrow gap against the original paper only when the README leaves one, and safe-debug diagnoses failures before anything gets patched. The explore lane only starts once you explicitly authorize it on top of a frozen research state, and explore-code and explore-run then implement and test candidate ideas on an isolated branch, writing ranked, no-overclaim results rather than a new best score.
This is built for a deep learning researcher reproducing someone else's repository or extending their own, who wants an audit trail (change logs, comparability notes, evidence ledgers) instead of an agent quietly patching a repo until a metric moves. Nine of the eleven skills are leaf or helper skills meant to be called by the two orchestrators above, not run directly, and none of the eleven do anything for work outside deep learning research.
READ THE FULL ANALYSIS
This is one person's framework, not an organization's. The repository is maintained under a single GitHub account, and its own README already documents one slug migration (ai-paper-reproduction to ai-research-reproduction) and a compatibility-fallback repo name — worth knowing before pointing a real research pipeline at it, and a reason to pin the exact skill slug you install.
Checked 19 September 2026 from the repository README and its 11 live skills.
WHAT'S INSIDE
11 showing · 11 totalReady to use
Nothing else to set up — install it and go.
11 of 11
SETUP
Showing all 11 skills
ai-research-explore
A separate track for trying new ideas on a research model that already works — each idea is tested on a copy, scored against the original, and reported as promising rather than proven.
ai-research-reproduction
Takes the code published with an AI research paper and gets it running on your own machine, keeping an honest record of what worked, what had to be changed, and what was guessed.
analyze-project
Reads through an AI project's code and explains how it fits together — where the model is defined, where training and prediction start, which settings matter — without changing or running a thing.
env-and-assets-bootstrap
The setup step before anything is run: which software environment to create, and where the downloaded model files, datasets and caches need to sit.
explore-code
Makes the actual code change behind an experiment — borrowing a component from another project, swapping out part of the model — on a copy, and writes down how to undo it.
explore-run
Runs the short, cheap trials of an experiment — a slice of the data, a few hundred training steps, a batch of settings at once — then ranks what came out and flags where the comparison isn't fair.
minimal-run-and-audit
You already know which command to run — this runs it and writes up what happened in a fixed format: worked, partly worked, or blocked, and which files had to be edited.
paper-context-resolver
When a project's instructions leave out the one detail you need to run it, this digs that single answer out of the paper behind it — and records where the paper and the code disagree.
repo-intake-and-plan
Opens a research project you have just downloaded and reports what is inside it, which commands its own documentation gives, and which of them is the smallest safe thing to try first.
run-train
Starts a model training job the careful way — a startup check first, then a short trial, then the real run — keeping the settings, logs and saved progress as proof of what happened.
safe-debug
Takes the error a failed training or prediction job printed, narrows down what actually caused it, and proposes the smallest fix; nothing in your code changes until you approve it.
No skill in this package matches that.
HOW TO GET IT
WHOLE PACKAGEInstall the whole package (11 skills)
npx skills add lllllllama/rigorpilot-skillsSINGLE SKILLOr install a single skill
npx skills add lllllllama/rigorpilot-skills --skill <name> --full-depthPick the skill name from the Skills tab — each entry there installs independently.